Community micro-spaces play a critical role in fostering social interactions and strengthening community cohesion. However, current designs of these spaces often overlook user behavioral patterns, resulting in inefficient space usage, low interactivity, and a lack of inclusivity. To address these issues and enhance the participation, inclusivity, and interactivity of community micro-spaces, this study introduces an interactive generative design and optimization framework based on behavioral performance. By integrating user behavioral data with generative design technologies, this framework aims to achieve precise optimization and personalized design for community micro-spaces. The study begins by outlining a five-stage workflow, which includes data collection, data processing, prompt engineering, space optimization, and experimental validation. Behavioral and perceptual data from micro-space users were gathered using PSPL surveys, Likert-scale perception questionnaires, and YOLO algorithms. These behavioral performance data were then converted into precise prompts and linked to LoRA models, exploring the relationship between user behavior and AIGC prompts. The Stable Diffusion platform was subsequently employed for space optimization. Finally, the proposed methods were applied to eight selected community micro-spaces to verify their effectiveness in improving space functionality and user experience. This research demonstrates the deep integration of behavioral data and generative AI in public space design, which significantly enhances the participatory process of community space updates. Additionally, it provides valuable theoretical and technical support for future urban space design, offering new insights into the application of AI-driven design optimization in public spaces.